AI Daily Digest

WEDNESDAY, MAY 6, 2026
Top Stories

OpenAI Makes GPT-5.5 Instant the Default ChatGPT Model

OpenAI rolled GPT-5.5 Instant out as the default model for ChatGPT, replacing GPT-5.3 Instant. The update targets reduced hallucinations in law, medicine, and finance while preserving low latency, with rollout to free, business, and enterprise tiers landing in the coming weeks.

Source: TechCrunch

Anthropic Doubles Claude Code Limits, Locks In SpaceX Colossus Capacity

Anthropic doubled the five-hour Claude Code window for Pro, Max, and Enterprise subscribers and signed for the entire capacity of SpaceX's Colossus data center — a substantial compute commitment as agentic coding workloads keep climbing.

Sierra Hits $15.8B Valuation in $950M Series E

Bret Taylor and Clay Bavor's agent-platform startup Sierra closed a $950M round led by Tiger Global and Google's GV. Sierra reportedly serves nearly half the Fortune 50 and runs at $150M ARR, anchoring the agent-infrastructure thesis driving 2026's largest AI checks.

Source: SiliconANGLE

Apple Settles Apple Intelligence / Siri Lawsuit for $250M

Apple agreed to a $250M class-action settlement in California federal court over allegations it marketed AI and Siri features that weren't yet shipping during the iPhone 15/16 launch. The settlement is the largest AI-marketing accountability event to date.

Source: NB SLA

Nvidia Agent Toolkit Launches with Adobe, Salesforce, SAP and 14 Other Adopters

At GTC 2026 Nvidia rolled out its Agent Toolkit, with Adobe positioning it as the foundation for hybrid long-running creativity and marketing agents. The 17-partner launch lineup signals Nvidia's push to own the enterprise agent runtime layer, not just the GPUs underneath.

Source: VentureBeat

AMD's Lisa Su Pins Massive Forecast Revision on Agentic AI Demand

AMD CEO Lisa Su told CNBC the company's outsized forecast bump is driven by a CPU demand surge tied to agentic AI workloads — a notable signal that inference is leaking off GPUs and into CPU territory as agent fleets scale.

Source: CNBC

DeepSeek-V4-Flash Sets New Intelligence-per-Parameter Standard

The DeepSeek-V4-Flash refresh (refined through May 4) uses a 284B-parameter MoE architecture with only 13B active per token, delivering frontier-level performance at a fraction of the compute cost. It joins a 12-day flurry of Chinese open-weights coding models from Z.ai, MiniMax, and Moonshot.

Source: devFlokers

Google DeepMind's TurboQuant Targets the KV Cache Bottleneck

DeepMind unveiled TurboQuant at ICLR 2026 — a quantization algorithm that significantly cuts the KV-cache memory overhead choking long-context inference. If it ships, expect cheaper long-context serving and longer effective windows on commodity hardware.

Source: Crescendo AI

Nippon Life v. OpenAI Raises Unauthorized-Practice-of-Law Claims

A new suit alleges that an OpenAI chatbot helped a user reopen a settled case and generated dozens of filings citing nonexistent authority. The case sharpens the liability question for general-purpose AI systems giving legal output without supervision.

Source: AI Business

Parag Agrawal's Parallel Web Systems Raises $100M from Sequoia

Former Twitter CEO Parag Agrawal closed a $100M round led by Sequoia for Parallel Web Systems. The deal joins a clear May pattern: capital flowing toward agent infrastructure and the substrate that lets AI act on the open web.

PIP Acquires AlphaAdvisors as Q1 AI M&A Volume Hits Record

Performance Improvement Partners closed its acquisition of AI/data advisory firm AlphaAdvisors on May 4. CB Insights logged 266 AI M&A deals in Q1 2026, a 90% YoY jump — momentum concentrated in infrastructure, advisory capacity, and platform consolidation.

IBM, Riken, and Cleveland Clinic Simulate 12,000-Atom Protein Complex

The team linked quantum systems with Japanese supercomputers to simulate a biologically meaningful protein complex. No quantum-advantage claim was made, but the work points at near-term utility for drug discovery — an early signal of hybrid quantum/AI research moving from demo to applied.

Quick Hits
ReFiBuy raises $13.6M seed — Raleigh startup helping e-commerce brands surface product data to ChatGPT and Claude shopping agents. Tech Startups
QuantWare raises $178M for superconducting quantum processors — Netherlands-based foundry pushing an open-architecture quantum hardware play. Tech Startups
Silicon Motion up 150% YTD — JPMorgan hiked its target from $145 to $260 after a blowout Q1; AI memory controllers are the lever. The Motley Fool
Micron forecasts ~400% earnings growth over two years — HBM demand for AI training continues to outstrip supply. The Motley Fool
CoreWeave projecting $10B+ 2026 revenue — neocloud serving Nvidia, OpenAI, Microsoft, and Meta keeps scaling. The Motley Fool
Canadian fiddler sues Google for $1.5M over AI Overview defamation — Ashley MacIsaac alleges AI Overview falsely tagged him as a sex offender. resultsense
Anthropic's Claude Mythos flagged for cybersecurity exploit capability — internal testing showed the model identifying and exploiting thousands of critical flaws in hardened environments. Crescendo AI
UPenn introduces "Mollifier Layers" for inverse PDEs — embeds classical smoothing into neural nets for stable scientific AI. blog.mean.ceo
Mimic Robotics shows 10x sample efficiency on real-world manipulation — video-action model with 2x faster convergence on physical tasks. blog.mean.ceo
Enterprise generative AI spend on track for $2.5B in 2026 — Gartner sees ~4x growth over 2025 as agent deployments hit 97% of executives' firms. globenewswire